Building Sales OS: AI Account Research That Ends in a Grade

AI Development

Our sales team was losing about forty minutes an account to one question: does this company own the equipment being inspected, or sell inspection as a service to somebody else.

One is a prospect. The other already sells what we would be helping them build. Dev Station built Sales OS to answer that from the domain alone, and every brief it returns ends in a grade from A+ down to D.

The Challenge

Sales OS began as an internal fix. Six problems in our own pipeline shaped it:

1. Research That Ended in a Document: Two careful pages on a prospect, filed, and nobody could say whether to call them this week.

2. The Deciding Fact Was Invisible: Whether a prospect owns inspection equipment or services equipment owned by somebody else almost never appears on a homepage.

3. Firmographics Filed Opposites Together: Same industry, same headcount, same region, and one runs a fleet of test rigs while the other sends engineers out to customer sites.

4. Forty Minutes an Account: The signal sits in service pages, job adverts and the equipment named in a careers post, all of it read by hand.

5. No Way to Check a Judgement: A gut call is often right, impossible to hand to a colleague, and impossible to review later.

6. Quiet Accounts Had No Trigger: An account that never replied stayed on the list with nothing to say whether it deserved a second attempt.

Our Solution

One Domain In, One Graded Brief Out: A run costs a single credit. The engine crawls the site, enriches through Apollo.io, verifies UK entities against Companies House, then synthesises the dossier with Claude.
Six Axes Behind the Grade: A composite out of 100 and a letter, split into Company Fit, Contact Access, Opportunity Quality, Competitive Position, Outreach Readiness and Why-Now. When a grade looks wrong, the breakdown shows which axis pulled it there.
The Buying Committee as Cards: Each person gets a persona, a fit score and the authority they hold, verified through Apollo. One card carries a START WITH marker.
Qualification With the Evidence Next to It: BANT and MEDDIC both get scored, with the note that produced each score. A list of what public sources could not establish turns the first call into a check of the gaps.
Outreach and Deal Prep: Three emails on day one, day four and day eight, written from what the crawl found, with merge tokens for any sequencer. Deal prep adds an opener, five discovery questions, the likely traps, and objection battlecards.
An Audit Trail That Doubles as a Buying Signal: Every dossier keeps its ID, workspace, date and engine version. Run the account again and the tool reports what changed on the site.
The Settings Page Sets Everything Else: Grades are only as sharp as your description of what you sell and who you sell to. Ticking manufacturing says almost nothing. Naming the operational pain says exactly what to look for.

Technology Stack

Backend: Python 3.13, FastAPI, Uvicorn, Pydantic 2
Database: SQLite by default, SQLAlchemy 2 with Alembic migrations, PostgreSQL configurable
Frontend: React 19, TypeScript, Vite 8
UI: Tailwind CSS 3, Radix UI, Lucide icons, Recharts
Frontend data: TanStack React Query 5
Marketing site: Next.js 15, React 19, TypeScript, Tailwind, Motion
Authentication: Clerk, with JWT verification in the backend
Payments: Stripe subscriptions, credits and webhooks
AI and sales data: Anthropic Claude compatible API, Apollo, Companies House
PDF export: ReportLab in the backend, jsPDF in the frontend
Testing and lint: pytest, Playwright, Oxlint on the frontend, ESLint on the marketing site
Deployment: Docker multi-stage build with Docker Compose

Key Outcomes & Impact

Research time: 40 accounts took about 26 hours of reading a week, now 40 credits and a ranked list
Every brief: a composite out of 100, a letter grade and a one line action
Six scored axes: an 82 with contact access at 70 reads as worth the week and hard to enter
Qualification: BANT at 25 points each, MEDDIC across 6 dimensions, every score carrying its evidence
One credit: opens all six tabs, account brief through to evidence and export
Cost control: provider cost and token count reported per workspace
Availability: free starter tier of 5 accounts a month, account brief open on every plan
Built by Dev Station: custom software for engineering, manufacturing and industrial teams in the UK and the US, covering inspection platforms, BLE and measurement device integration, offline-first field apps, and the data layer behind them

Our Case Study

Nothing speaks louder than results. Explore how Dev Station Technology has helped other enterprises transform their ideas into digital products with outstanding user experiences and clear business impact.